CN108763319B - Social robot detection method and system fusing user behaviors and text information - Google Patents
Social robot detection method and system fusing user behaviors and text information Download PDFInfo
- Publication number
- CN108763319B CN108763319B CN201810396753.7A CN201810396753A CN108763319B CN 108763319 B CN108763319 B CN 108763319B CN 201810396753 A CN201810396753 A CN 201810396753A CN 108763319 B CN108763319 B CN 108763319B
- Authority
- CN
- China
- Prior art keywords
- vector
- behavior
- text
- user
- feature vector
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000006399 behavior Effects 0.000 title claims abstract description 187
- 238000001514 detection method Methods 0.000 title claims abstract description 63
- 239000013598 vector Substances 0.000 claims abstract description 352
- 238000000034 method Methods 0.000 claims abstract description 54
- 239000011159 matrix material Substances 0.000 claims description 56
- 238000006243 chemical reaction Methods 0.000 claims description 42
- 238000013528 artificial neural network Methods 0.000 claims description 39
- 230000000306 recurrent effect Effects 0.000 claims description 28
- 230000004927 fusion Effects 0.000 claims description 24
- 238000013527 convolutional neural network Methods 0.000 claims description 22
- 230000006870 function Effects 0.000 claims description 18
- 238000005295 random walk Methods 0.000 claims description 18
- 238000004422 calculation algorithm Methods 0.000 claims description 16
- 238000013507 mapping Methods 0.000 claims description 16
- AGVAZMGAQJOSFJ-WZHZPDAFSA-M cobalt(2+);[(2r,3s,4r,5s)-5-(5,6-dimethylbenzimidazol-1-yl)-4-hydroxy-2-(hydroxymethyl)oxolan-3-yl] [(2r)-1-[3-[(1r,2r,3r,4z,7s,9z,12s,13s,14z,17s,18s,19r)-2,13,18-tris(2-amino-2-oxoethyl)-7,12,17-tris(3-amino-3-oxopropyl)-3,5,8,8,13,15,18,19-octamethyl-2 Chemical compound [Co+2].N#[C-].[N-]([C@@H]1[C@H](CC(N)=O)[C@@]2(C)CCC(=O)NC[C@@H](C)OP(O)(=O)O[C@H]3[C@H]([C@H](O[C@@H]3CO)N3C4=CC(C)=C(C)C=C4N=C3)O)\C2=C(C)/C([C@H](C\2(C)C)CCC(N)=O)=N/C/2=C\C([C@H]([C@@]/2(CC(N)=O)C)CCC(N)=O)=N\C\2=C(C)/C2=N[C@]1(C)[C@@](C)(CC(N)=O)[C@@H]2CCC(N)=O AGVAZMGAQJOSFJ-WZHZPDAFSA-M 0.000 claims description 9
- 229940029329 intrinsic factor Drugs 0.000 claims description 9
- 230000009466 transformation Effects 0.000 claims description 8
- 238000012549 training Methods 0.000 claims description 7
- 238000011176 pooling Methods 0.000 claims description 6
- 239000000126 substance Substances 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 2
- 230000015654 memory Effects 0.000 description 11
- 230000008569 process Effects 0.000 description 7
- 238000010586 diagram Methods 0.000 description 5
- 125000004122 cyclic group Chemical group 0.000 description 4
- 239000010410 layer Substances 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 230000007787 long-term memory Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000006403 short-term memory Effects 0.000 description 3
- 230000003542 behavioural effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 239000013604 expression vector Substances 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 239000002356 single layer Substances 0.000 description 2
- 238000007619 statistical method Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000003245 working effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Machine Translation (AREA)
Abstract
The invention belongs to the technical field of computers, and particularly provides a social robot detection method and system fusing user behaviors and text information. The method for detecting the social robot aims to solve the problems that in the prior art, characteristics are selected manually, logicality and time sequence among social media posts are ignored, and behavior information of a social platform user is ignored; obtaining a user text feature vector, a behavior feature vector and a friend network feature vector based on the data, and fusing the user text feature vector, the behavior feature vector and the friend network feature vector to obtain a user feature vector of the social media user to be detected; and detecting the user characteristic vector and outputting a detection result. The method provided by the invention better conforms to the characteristics of the social media, and the social media user to be detected is analyzed from multiple dimensions, so that the detection accuracy is improved. The system of the present invention also has the above-described advantageous effects.
Description
Technical Field
The invention belongs to the technical field of computers, and particularly relates to a social robot detection method and system fusing user behaviors and text information.
Background
With the rapid development of internet technology and social media platforms, a large number of social robots exist in various mainstream social media platforms at home and abroad, the social robots make and release false messages, propagate rumors, make network traps, influence or even control public opinion development trends, seriously interfere with the normal life of net citizens, threaten the privacy safety of social media users, and even cause bad influence on the society, so how to accurately detect the robots in the social media platforms, prevent negative influence brought by the social robots, and have very important practical value.
The early social robot identification mainly depends on a specific manual strategy, and the social robot takes the fact that the self influence of the social robot is improved through wide friend making as a starting point, a large number of detection seed account numbers are randomly manufactured, the account numbers have no actual behaviors and do not release meaningful contents, so that a human user does not establish a friendly relationship with the detection seed account numbers, and finally the social robot is detected from the account numbers establishing the friendly relationship with the seed account numbers through a series of rules, but the method is original and simple, needs to consume more manpower and time, and cannot be well applied to practice; secondly, an analysis method of the social robot based on network structure analysis is provided, the method is characterized in that the junk account is supposed to be connected with only a few real users, most of the rest are the junk accounts, and the junk accounts which are densely connected are identified through the characteristics; for the text information of social media, a language feature-based correlation method is proposed, which statistically analyzes the average length of the published text information, the average length of the URL (Uniform resource Locator) in the text, and the like by analyzing specific words and punctuation marks.
The social robot detection method in the prior art mainly has the following problems: 1. based on the existing statistical method and machine learning method, the characteristics need to be selected manually, and a large amount of labor cost is consumed; 2. the existing social robot detection method treats the content published by a user as pure text information, and ignores the logic and time sequence among social media posts; 3. the prior art method often ignores the behavior information of the user in the social platform, or only adopts a simple statistical method for analysis, and cannot effectively analyze and utilize the user behavior information in the social platform.
Therefore, how to propose a solution to the above problems is a problem that needs to be solved by those skilled in the art.
Disclosure of Invention
In order to solve the above problems in the prior art, the present invention provides a social robot detection method fusing user behavior and text information, including:
acquiring historical network data and friend network data of a social media user to be detected;
converting the historical network data into a user text feature vector based on a pre-constructed first vector conversion model;
converting the historical network data into behavior feature vectors based on a pre-constructed second vector conversion model;
converting the friend network data into friend network feature vectors based on a pre-constructed third vector conversion model;
fusing the user text characteristic vector, the behavior characteristic vector and the friend network characteristic vector to obtain a user characteristic vector of the social media user to be detected;
detecting the user characteristic vector based on a pre-constructed classification detection model, and outputting a detection result;
wherein the content of the first and second substances,
the first vector transformation model, the second vector transformation model, the third vector transformation model and the classification detection model are all models which are constructed on the basis of a preset training set and by utilizing a deep neural network.
In a preferred technical solution of the above method, the historical network data includes text data, and the step of converting the historical network data into a user text feature vector includes:
mapping the text data to a text matrix sequence based on a word vector model;
encoding the text matrix sequence into a text feature vector sequence by using a convolutional neural network;
and encoding the text feature vector sequence into a user text feature vector by using a recurrent neural network.
In a preferred embodiment of the above method, the step of encoding the text matrix sequence into a text feature vector sequence by using a convolutional neural network includes:
carrying out convolution operation on the text matrix sequence by utilizing the convolution layer of the convolution neural network to obtain a characteristic mapping matrix, wherein the method is shown in the following formula:
clk＝(S*Fl)k＝∑mω(S[:,k-m+1:k]⊙Fl)mω
wherein S represents the text matrix sequence, FlRepresenting the filter, representing the convolution operation, m representing the width of the filter, S[:,k-m+1:k]Representing an m-dimensional matrix slice, ω representing the length of the text data, and k representing an intermediate variable;
and performing pooling operation on the feature mapping matrix by using a pooling layer of the convolutional neural network to obtain a text feature vector sequence.
In the preferred technical solution of the above method, the step of "encoding the text feature vector sequence into a user text feature vector using a recurrent neural network" is shown by the following formula:
it＝σ(Wi[ht-1,twtt]+bi)
ft＝σ(Wf[ht-1,twtt]+bf)
qt＝tanh(Wc[ht-1,twtt]+bc)
ot＝σ(Wo[ht-1,twtt]+bo)
ct＝ft⊙ct-1+it⊙qt
ht＝ot⊙tanh(ct)
wherein itDenotes an input gate, ftIndicating a forgetting gate, qtCandidate vectors representing text data entered at time t, ctRepresenting the state of the cells of the recurrent neural network at time t, otDenotes the output gate, htIndicating an implicit state at time t, i.e. output information, Wi，Wf，Wc，Wo，bi，bf，bc，boAre learning parameters of the recurrent neural network, sigma (·) represents a sigmoid function, tanh (·) represents a hyperbolic tangent function, twttRepresenting the text feature vector.
In a preferred embodiment of the foregoing method, the historical network data includes behavior data, and the step of converting the historical network data into a behavior feature vector includes:
constructing an internal factor behavior modeling component and an external factor behavior modeling component;
encoding the behavior data into an intrinsic behavior vector based on the intrinsic factor behavior modeling component;
encoding the intrinsic behavior vector as an extrinsic behavior vector based on the extrinsic factor behavior modeling component;
encoding the extrinsic behavior vectors into behavior feature vectors through a recurrent neural network, wherein the behavior feature vectors include original behavior feature vectors and forwarding behavior feature vectors.
In the preferred technical solution of the above method, the step of converting the friend network data into friend network feature vectors includes:
generating a random walk sequence corresponding to the friend network data by using a random walk algorithm;
the random walk sequence is encoded into a friend network feature vector using the Skip-Gram algorithm.
In the preferred technical solution of the above method, the step of fusing the user text feature vector, the behavior feature vector, and the friend network feature vector includes:
and serially splicing the user text characteristic vector, the behavior characteristic vector and the friend network characteristic vector, wherein the method is shown in the following formula:
Uu＝[UCu,Pru,Rru,ctu]
wherein, UCuRepresenting said text feature vector, PruRepresenting the feature vector of the original behavior, RruA feature vector, ct, representing the forwarding behavioruRepresenting the friend network feature vector.
In the preferred technical solution of the above method, the step of fusing the user text feature vector, the behavior feature vector, and the friend network feature vector includes:
fusing the user text characteristic vector, the behavior characteristic vector and the friend network characteristic vector based on a preset weight matrix, wherein the method is shown in the following formula:
Uu＝B+(Pru+V·Rru)+Wc·UCu+Wn·ctu
wherein B represents the global bias, V represents the weight for balancing the original behavior feature vector with the forwarding behavior feature vector, WcWeights, W, representing the text feature vectorsnWeights representing the friend network feature vectors.
In the preferred technical solution of the above method, the method of "detecting the user text feature vector based on the pre-constructed classification detection model" is as follows:
wherein H and H represent the weight matrix and bias of the classification detection model, UuRepresents the user feature vector, σ (-) represents a sigmoid function,
indicating the detection result.
The second aspect of the present invention also provides a social robot detection system fusing user behavior and text information, including:
the acquisition module is configured to acquire historical network data and friend network data of a social media user to be detected;
a first vector conversion module configured to convert the historical network data into a user text feature vector;
a second vector conversion module configured to convert the historical network data into behavior feature vectors;
the third vector conversion module is configured to convert the friend network data into friend network feature vectors;
the fusion module is configured to fuse the user text feature vector, the behavior feature vector and the friend network feature vector to obtain a user feature vector of the to-be-detected social media user;
the classification detection module is configured to detect the user feature vector and output a detection result;
the first vector conversion module, the second vector conversion module, the third vector conversion module, the fusion module and the classification detection module are all modules constructed on the basis of a preset training set and by utilizing a deep neural network.
In a preferred technical solution of the above scheme, the historical network data includes text data, and the first vector conversion module further includes a mapping unit, a convolutional neural network unit, and a cyclic neural network unit;
the mapping unit is configured to map the text data into a text matrix sequence based on a word vector model;
the convolutional neural network unit is configured to encode the text matrix sequence into a text feature vector sequence by using a convolutional neural network;
the recurrent neural network unit is configured to encode the sequence of text feature vectors into a user text feature vector using a recurrent neural network.
In a preferred technical solution of the above scheme, the convolutional neural network unit is further configured to perform a convolution operation on the text matrix sequence according to a formula shown as follows:
clk＝(S*Fl)k＝∑mω(S[:,k-m+1:k]⊙Fl)mω
wherein S represents the text matrix sequence, FlRepresenting the filter, representing the convolution operation, m representing the width of the filter, S[:,k-m+1:k]Represents an m-dimensional matrix slice, ω represents the length of the text data, and k represents an intermediate variable.
In a preferred technical solution of the above scheme, the recurrent neural network unit is further configured to encode the text feature vector sequence into a user text feature vector according to a formula shown below:
it＝σ(Wi[ht-1,twtt]+bi)
ft＝σ(Wf[ht-1,twtt]+bf)
qt＝tanh(Wc[ht-1,twtt]+bc)
ot＝σ(Wo[ht-1,twtt]+bo)
ct＝ft⊙ct-1+it⊙qt
ht＝ot⊙tanh(ct)
wherein itDenotes an input gate, ftIndicating a forgetting gate, qtCandidate vectors representing text data entered at time t, ctRepresenting the state of the cells of the recurrent neural network at time t, otDenotes the output gate, htIndicating an implicit state at time t, i.e. output information, Wi，Wf，Wc，Wo，bi，bf，bc，boAre learning parameters of the recurrent neural network, sigma (·) represents a sigmoid function, tanh (·) represents a hyperbolic tangent function, twttRepresenting the text feature vector.
In a preferred technical solution of the foregoing scheme, the historical network data includes behavior data, and the second vector conversion module further includes a behavior modeling component unit, a first encoding unit, a second encoding unit, and a third encoding unit;
the behavior modeling component unit is configured to construct an intrinsic factor behavior modeling component and an extrinsic factor behavior modeling component;
the first encoding unit is configured to encode the behavior data into an intrinsic behavior vector based on the intrinsic factor behavior modeling component;
the second encoding unit is configured to encode the intrinsic behavior vector as an extrinsic behavior vector based on the extrinsic factor behavior modeling component;
the third encoding unit is configured to encode the extrinsic behavior vectors into behavior feature vectors through a recurrent neural network, wherein the behavior feature vectors include original behavior feature vectors and forwarding behavior feature vectors.
In a preferred technical solution of the above scheme, the third vector conversion module includes: a sequence generating unit and a fourth encoding unit;
the sequence generating unit is configured to generate a random walk sequence corresponding to the friend network data by using a random walk algorithm;
the fourth encoding unit is configured to encode the random walk sequence into a friend network feature vector using a Skip-Gram algorithm.
In a preferred technical solution of the above scheme, the fusion module further includes a first fusion unit, and the first fusion unit is configured to concatenate the user text feature vector, the behavior feature vector, and the friend network feature vector according to the following formula:
Uu＝[UCu,Pru,Rru,ctu]
wherein, UCuRepresenting said user text feature vector, PruRepresenting the feature vector of the original behavior, RruA feature vector, ct, representing the forwarding behavioruRepresenting the friend network feature vector.
In a preferred technical solution of the foregoing solution, the fusion module further includes a second fusion unit, and the second fusion unit is configured to fuse the user text feature vector, the behavior feature vector, and the friend network feature vector based on a preset weight matrix according to the following formula:
Uu＝B+(Pru+V·Rru)+Wc·UCu+Wn·ctu
where B denotes the global bias, V denotes the weight used to balance the originating behavior vector with the forwarding behavior vector, WcWeights, W, representing the user text feature vectorsnWeights representing the friend network feature vectors.
In a preferred technical solution of the above scheme, the classification detection module is configured to detect the user text feature vector according to the following formula:
wherein H and H represent the weight matrix and bias of the classification detection model, UuRepresents the user feature vector, σ (-) represents a sigmoid function,
indicating the detection result.
Compared with the closest prior art, the invention provides a social robot detection method fusing user behaviors and text information, which comprises the steps of obtaining historical network data and friend network data of a social media user to be detected; converting historical network data into a user text feature vector based on a pre-constructed first vector conversion model, and converting the historical network data into a behavior feature vector based on a pre-constructed second vector conversion model; converting friend network data into friend network feature vectors based on a pre-constructed third vector conversion model; fusing the user text characteristic vector, the behavior characteristic vector and the friend network characteristic vector to obtain a user characteristic vector of the social media user to be detected; detecting the user characteristic vectors based on a pre-constructed classification detection model, and outputting a detection result;
the technical scheme at least has the following beneficial effects:
1. according to the method, the text data of the historical network data is converted into the text feature vector of the user based on the constructed vector conversion model, so that the time sequence relation and the logic relation between social media texts can be mined, and the characteristics of the social media are better met;
2. the method analyzes the behavior characteristics of the user according to the internal factors and the external factors, mines and analyzes the behavior pattern of the user, integrates text information, behavior information and friend network information, analyzes the social media user to be detected from multiple dimensions, and improves the detection accuracy of the social robot.
Drawings
Fig. 1 is a schematic diagram illustrating main steps of a social robot detection method fusing user behaviors and text information according to an embodiment of the present invention;
FIG. 2 is a diagram illustrating a learning process of a user text feature vector according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating a learning process of a user behavior feature vector according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of a social robot detection system fusing user behaviors and text information according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, are within the scope of the present invention
Preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only for explaining the technical principle of the present invention, and are not intended to limit the scope of the present invention.
Referring to fig. 1, fig. 1 exemplarily shows main steps of a social robot detection method for fusing user behavior and text information in the present embodiment. As shown in fig. 1, the social robot detection method fusing user behavior and text information in this embodiment includes the following steps:
step S101: and acquiring historical network data and friend network data of the social media user to be detected.
Specifically, the historical network data at least comprises text data, release timestamp data and release behavior type data, in the embodiment of the invention, the release timestamp data is uniformly converted into Unix timestamps, namely the number of seconds from 1/1970 to the time of obtaining the release timestamp data, the release behavior type data mainly comprises original behavior type data and forwarding behavior type data, wherein the specific release behavior type data of the social media user to be detected is related to the social media platform, and the specific release behavior type data comprises but is not limited to the two types.
Step S102: and coding the text data based on the historical network data to obtain the text characteristic vector of the social media user to be detected.
As shown in fig. 2, fig. 2 exemplarily shows a learning process of a user text feature vector, and specifically, obtaining the user text feature vector may include the following steps:
step S1021: text vectorization of text data of the historical network data.
Taking historical data issued by the social media user to be detected as the post, the post S issued by the social media user to be detected is taken as an exampleiSequences [ W ] viewed as a series of words1,W2,…,Wl]Where l is the maximum length of the post and W represents a particular word. For post SiDefining a standard length omega, and for the text exceeding the standard length, cutting off the part of the standard length, namely when omega<When l, post SiIs represented by [ W ]1,W2,…,Wω](ii) a For text shorter than the standard length, the space is filled up with spaces, and the spaces are marked null, namely when omega>When l, post SiIs represented by [ W ]1,…,Wl,null,…,null]. If the text language of the historical data issued by the social media user to be detected is English, the processing is directly carried out, if the text language is Chinese, the historical data is subjected to word segmentation preprocessing, and then the processing is carried out, so that the prior art of the part is already disclosed, and the expansion description is not carried out.
Specifically, the step of converting the word sequence into the word vector may be:
using word vector model to convert word sequence wiVector x mapping into an e-dimensional vector spaceiVector xiI.e. the word vector of the word, wherein the dimension of e may be 100 or 200, and the specific value may be selected according to the actual situation. For post SiSequentially arranging the words w in the postsiConversion to word vector xiThat is, a text matrix of posts can be obtained
For a specified social media user u, a sequence [ S ] containing n historical post texts can be obtained by the methodu1,Su2,…,Sun]By adopting a word vector method, a text rectangle corresponding to each post can be obtained, namely a text matrix sequence [ S ] can be obtainedu1,Su2,…,Sun]。
Step S1022: and calculating the feature vector of the text matrix.
Specifically, in the embodiment of the present invention, the first vector transformation model may be a model based on a combination of a convolutional neural network and a cyclic neural network, and for convenience of description, the present invention is described by taking the convolutional neural network and the cyclic neural network as examples.
Inputting the text matrix sequence calculated in step S1021 into a convolutional neural network, the convolutional layer of which comprises S filters
Wherein m represents the width of the filter, and the text matrix sequence is convolved with the filter to map the text matrix into vectors
Wherein, the k element clkThe specific calculation process of (a) is shown in the following formula (1):
clk＝(S*Fl)k＝∑mω(S[:,k-m+1:k]⊙Fl)mω (1)
wherein, denotes convolution, S[:,k-m+1:k]Representing an m-dimensional matrix slice. When all convolution operations are finished, the feature mapping matrix can be obtained
Then inputting the characteristic mapping matrix C into the pooling layer of the convolutional neural network, and adopting a maximum pooling algorithm, namely for each row vector C of the matrix ClTaking the maximum value of its element to represent clFinally, the matrix C can be converted into eigenvectors
I.e. the feature vector of the text matrix S.
Step S1023: and constructing a text matrix feature vector sequence.
Text based on step S1021The matrix sequence and the feature vector of the text matrix obtained in step S1022 construct a text feature vector sequence SC of the historical data of the social media user to be detectedu＝[twtu1,twtu2,…,twtun]。
Step S1024: and analyzing the time sequence relation of different historical data of the social media user to be detected to obtain the text expression vector of the user.
Specifically, the recurrent neural network for processing natural language can be adopted to analyze the time sequence relation of different posts of the same user, the social media text is different from the traditional text, the social media text has the great characteristic of time sequence, the social media text contains time sequence information, posts sent by the same person within a period of time have certain correlation, and the detection effect of the social robot can be improved by mining the time sequence information.
In the embodiment of the present invention, the step is described by taking the long-short term memory network as an example, but it should be understood by those skilled in the art that the embodiment of the present invention is not limited to this model. The long-short term memory network mainly comprises three important components, namely a forgetting gate, an input gate and an output gate, wherein the forgetting gate determines which historical information is reserved, the input gate determines which newly input information is received, and the output gate determines which information in the current state is output as a result.
The text feature vector sequence SC obtained in step S1023u＝[twtu1,twtu2,…,twtun]Inputting the long and short term memory network, and calculating to obtain the user text expression vector UCuThe text representation vector may represent the text information of all historical posts of user u and the time-sequence relationship between them. Specifically, at time t, the components and states in the long-short term memory network are expressed by the following equations (2) to (7):
it＝σ(Wi[ht-1,twtt]+bi) (2)
ft＝σ(Wf[ht-1,twtt]+bf) (3)
qt＝tanh(Wc[ht-1,twtt]+bc) (4)
ot＝σ(Wo[ht-1,twtt]+bo) (5)
ct＝ft⊙ct-1+it⊙qt (6)
ht＝ot⊙tanh(ct) (7)
wherein itDenotes an input gate, ftIndicating a forgetting gate, qtCandidate vectors representing newly input text information at time t, ctRepresenting the cell state at time t of the long-short term memory network, otDenotes the output gate, htIndicating the hidden state at time t, i.e. the current output information, Wi，Wf，Wc，Wo，bi，bf，bc，boAll represent parameters required to be learned by the long-short term memory network, sigma (DEG) represents a sigmoid function, tanh (DEG) represents a hyperbolic tangent function, and when the time t is equal to n, the long-short term memory network outputs hnFinally, the user text feature vector UC is obtainedu＝hnAnd the method is used for representing the text information of all historical posts of the user u and the time sequence information among the historical posts.
Step S103: and calculating the user behavior feature vector.
As shown in fig. 3, fig. 3 exemplarily shows a schematic diagram of a behavior feature vector learning process, and in practical applications, a user behavior is mainly influenced by two factors, one is influenced by an internal factor, and the other is influenced by an external factor.
The intrinsic factors have important influence in the daily life of the user, the most direct expression is the biological clock, for example, some users are used to get up in the morning and use a mobile phone to surf the internet and make comments on social media, some users are used to use the social media before sleeping at night, and different users have larger difference in behavior patterns shown on a social media platform due to the influence of the intrinsic factors; the external factors are more derived from the influence of life, culture and environment on the user, such as weekends and holidays, the user often spends more time on the social media platform, the possibility of publishing new content is higher, and the time period and frequency of using the social media platform are greatly different for users with different working properties. In addition, for hot events of certain features, the relevant users may be more interested in and parameter into social media interactions of the relevant events during the time period of the feature.
Calculating the user behavior feature vector may comprise the steps of:
step S1031: the behavioral patterns of the user are analyzed using an intrinsic factor behavior modeling component.
Specifically, 24 hours are divided into unit time periods of fine granularity, which are recorded as (1,2, …, T), and in the unit time periods, behavior data of users, such as number of posts and the like, are counted and taken as feature values of the time periods, which are recorded as numa,iWherein a ═ p (original) or a ═ r (forward), i ═ 1,2, …, T, constructing a behavior information vector specifying a user within 24 hours, wherein the intrinsic behavior vector includes an intrinsic original behavior vector and an intrinsic forward behavior vector, and the intrinsic original behavior vector p is the intrinsic original behavior vectorud＝(nump,1,nump,2,…,nump,T) Intrinsic forwarding behavior vector rud＝(numr,1,numr,2,…,numr,T) And analyzing the behavior information vector, and mining the influence of the internal factors on the social media user behavior model.
Step S1032: the behavioral patterns of the user are analyzed using an extrinsic factor behavior modeling component.
Specifically, on the basis of the behavior data of the user in one day, the behavior pattern of the user in a period of time (month, year and the like) is analyzed and specified, and behavior fluctuation caused by various external factors (such as holidays, emergencies and the like) is analyzed and modeled. Taking the time period as D days as an example, for the original release behavior, p corresponding to the D day is calculated according to the step S1031udValues where D ═ 1,2, …, D. Integrating to obtain a behavior information vector sequence Pu＝[pu1,pu2,…,puD]. For theForwarding behavior, obtaining behavior information vector sequence R according to the same methodu＝[ru1,ru2,…,ruD]A behavior information vector PuAnd RuInputting the long and short term memory network to finally obtain the user behavior characteristic vectors of PruAnd Rru。
With PuFor example, at time t (t ═ 1,2, …, D), the components and states in the long-short term memory network are shown in equations (8) to (13) below:
it＝σ(Wi[ht-1,put]+bi) (8)
ft＝σ(Wf[ht-1,put]+bf) (9)
qt＝tanh(Wc[ht-1,put]+bc) (10)
ot＝σ(Wo[ht-1,put]+bo) (11)
ct＝ft⊙ct-1+it⊙qt (12)
ht＝ot⊙tanh(ct) (13)
wherein itDenotes an input gate, ftIndicating a forgetting gate, qtCandidate vector representing new input behavior information at time t, ctRepresenting the cell state at time t of the long-short term memory network, otDenotes the output gate, htAnd the implicit state at the moment t is represented, namely the current output information. Wi，Wf，Wc，Wo，bi，bf，bc，boAre the parameters that the model needs to learn. σ (-) denotes a sigmoid function, and tanh (-) denotes a hyperbolic tangent function. When the time t is equal to D, the long-short term memory network outputs hDFinally, we obtain the characteristic vector Pr of the original posting behavior of the useru＝hD. In the same way, the RuInputting the long and short term memory network, and obtaining the output of the network at the time D, namely obtaining the characteristic vector Rr of the user forwarding behavioru。
Step S104: and learning the friend network characteristic vector of the user based on the friend network data of the user.
Specifically, the network vector problem can be converted into a word vector problem in natural language processing by using a Deepwalk algorithm, for a friend network, one path is regarded as a sentence, and a node on the path is regarded as a word in the sentence, that is, the calculation problem of the network node vector can be equivalent to the word vector problem. A Random Walk sequence, i.e., the above path, is generated by a Random Walk (Random Walk) algorithm. In the word vector algorithm, the ith word is predicted according to the front and rear w words. By taking this thinking into account, the ith node, namely represented as P (v), is predicted by front and back w nodes by using Skip-Gram algorithmi|Φ(vi-w),…,Φ(vi-1),Φ(vi+1),…,Φ(vi+w) Where Φ is the mapping function.
Step S105: and fusing the user text feature vector, the behavior feature vector and the friend network feature vector to construct a user feature vector.
Specifically, the fusion method may include at least two methods:
the method comprises the following steps: the method for splicing the user text characteristic vector, the behavior characteristic vector and the friend network characteristic vector in series is utilized to splice the user text characteristic vector, the behavior characteristic vector and the friend network characteristic vector in series, and the specific method is as shown in the following formula (14):
Uu＝[UCu,Pru,Rru,ctu] (14)
wherein, UCuRepresenting the feature vector of the user text, PruRepresenting the original behavior vector, RruRepresents a forwarding behavior vector, ctuRepresenting a friend network feature vector.
The second method comprises the following steps: the user text characteristic vector, the behavior characteristic vector and the friend network characteristic vector are fused by setting and training a weight matrix, and the specific method is shown in the following formula (15):
Uu＝B+(Pru+V·Rru)+Wc·UCu+Wn·ctu (15)
wherein, WcWeights representing user text feature vectors; wnWeights representing network characteristics; pr (Pr) ofu+V·RruRepresenting behavior information, V is used for balancing weight between originality and forwarding; b represents the overall deviation. Wc，WnAnd V and B four weight matrixes are obtained through training.
Step S106: and constructing a classifier and detecting the social robot.
Specifically, a user feature vector U of a social media user U is trained through a single-layer perceptron networkuInputting a trained single-layer perceptron network and inputting a final detection and identification result of the social robot, wherein the specific method is shown in the following formula (16):
where H and H represent the weight matrix and bias of the classification detection model, UuRepresenting the user feature vector, sigma (-) represents the sigmoid function,
indicating the detection result.
The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied in hardware, a software module executed by a processor, or a combination of the two. A software module may reside in Random Access Memory (RAM), memory, Read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
Based on the embodiment of the detection method of the social robot, the invention also provides a social robot detection system fusing the user behavior and the text information. The social robot detection system fusing user behaviors and text information is described below with reference to the accompanying drawings.
Referring to fig. 4, fig. 4 illustrates a main structure of a social robot detection system that merges user behavior and text information in the present embodiment. As shown in fig. 4, the social robot detection system fusing user behavior and text information according to this embodiment includes an obtaining module 1, a first vector conversion module 2, a second vector conversion module 3, a third vector conversion module 4, a fusing module 5, and a classification detection module 6.
The system comprises an acquisition module 1, wherein the acquisition module 1 is configured to acquire historical network data and friend network data of a social media user to be detected;
the first vector conversion module 2, the first vector conversion module 2 is configured to convert the historical network data into a user text feature vector;
the second vector conversion module 3, the second vector conversion module 3 is configured to convert the historical network data into behavior feature vectors;
the third vector conversion module 4 is configured to convert the friend network data into friend network feature vectors;
the fusion module 5 is configured to fuse the user text feature vector, the behavior feature vector and the friend network feature vector to obtain a user feature vector of the social media user to be detected;
the classification detection module 6 is configured to detect the user feature vectors and output detection results;
the first vector conversion module 2, the second vector conversion module 3, the third vector conversion module 4, the fusion module 5 and the classification detection module 6 are all modules constructed based on a preset training set and by utilizing a deep neural network.
In a preferred technical solution of the above scheme, the historical network data includes text data, and the first vector conversion module 2 includes a mapping unit, a convolutional neural network unit, and a cyclic neural network unit;
the mapping unit is configured to map the text data into a text matrix sequence based on the word vector model;
the convolutional neural network unit is configured to encode the text matrix sequence into a text feature vector sequence by using a convolutional neural network;
the recurrent neural network unit is configured to encode the sequence of text feature vectors into user text feature vectors using a recurrent neural network.
In a preferred embodiment of the foregoing solution, the convolutional neural network unit is further configured to perform a convolution operation on the text matrix sequence according to formula (1).
In a preferred technical solution of the above scheme, the recurrent neural network unit is further configured to encode the text feature vector sequence into the user text feature vector according to formulas (2) - (7).
In a preferred technical solution of the above scheme, the historical network data includes behavior data, and the second vector conversion module 3 further includes a behavior modeling component unit, a first encoding unit, a second encoding unit, and a third encoding unit;
the behavior modeling component unit is configured to construct an intrinsic factor behavior modeling component and an extrinsic factor behavior modeling component;
the first encoding unit is configured to encode the behavior data into an intrinsic behavior vector based on the intrinsic factor behavior modeling component;
the second encoding unit is configured to encode the intrinsic behavior vector into an extrinsic behavior vector based on the extrinsic factor behavior modeling component;
the third encoding unit is configured to encode the extrinsic behavior vectors into behavior feature vectors through a recurrent neural network, wherein the behavior feature vectors include original behavior feature vectors and forwarding behavior feature vectors.
In a preferred technical solution of the above scheme, the third vector conversion module 4 includes a sequence generation unit and a fourth encoding unit;
the sequence generation unit is configured to generate a random walk sequence corresponding to the friend network data by using a random walk algorithm;
the fourth encoding unit is configured to encode the random walk sequence into a friend network feature vector using a Skip-Gram algorithm.
In a preferred technical solution of the above scheme, the fusion module 5 further includes a first fusion unit, and the first fusion unit is configured to concatenate the user text feature vector, the behavior feature vector, and the friend network feature vector according to a formula (14).
In a preferred technical solution of the above scheme, the fusion module 5 further includes a second fusion unit, and the second fusion unit is configured to fuse the user text feature vector, the behavior feature vector, and the friend network feature vector according to a formula (15) based on a preset weight matrix.
In a preferred embodiment of the foregoing solution, the classification detection module 6 is configured to detect the user feature vector according to formula (16).
It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working process and the related descriptions of the system according to the embodiment of the present invention may refer to the corresponding process in the method according to the foregoing embodiment, and have the same beneficial effects as the method described above, and are not repeated herein.
Those of skill in the art will appreciate that the method steps of the examples described in connection with the embodiments disclosed herein may be embodied in electronic hardware, computer software, or combinations of both, and that the components and steps of the examples have been described above generally in terms of their functionality in order to clearly illustrate the interchangeability of electronic hardware and software. Whether such functionality is implemented as electronic hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing or implying any particular order or sequence. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein.
So far, the technical solutions of the present invention have been described in connection with the preferred embodiments shown in the drawings, but it is easily understood by those skilled in the art that the scope of the present invention is obviously not limited to these specific embodiments. Equivalent changes or substitutions of related technical features can be made by those skilled in the art without departing from the principle of the invention, and the technical scheme after the changes or substitutions can fall into the protection scope of the invention.
Claims (16)
1. A social robot detection method fusing user behaviors and text information is characterized by comprising the following steps:
acquiring historical network data and friend network data of a social media user to be detected;
converting the historical network data into a user text feature vector based on a pre-constructed first vector conversion model;
converting the historical network data into behavior feature vectors based on a pre-constructed second vector conversion model;
converting the friend network data into friend network feature vectors based on a pre-constructed third vector conversion model;
fusing the user text characteristic vector, the behavior characteristic vector and the friend network characteristic vector to obtain a user characteristic vector of the social media user to be detected;
detecting the user characteristic vector based on a pre-constructed classification detection model, and outputting a detection result;
wherein the content of the first and second substances,
the first vector transformation model, the second vector transformation model, the third vector transformation model and the classification detection model are all models which are constructed on the basis of a preset training set and by utilizing a deep neural network;
the historical network data comprises behavior data, and the step of converting the historical network data into behavior feature vectors comprises the following steps:
constructing an internal factor behavior modeling component and an external factor behavior modeling component;
encoding the behavior data into an intrinsic behavior vector based on the intrinsic factor behavior modeling component;
encoding the intrinsic behavior vector as an extrinsic behavior vector based on the extrinsic factor behavior modeling component;
encoding the extrinsic behavior vectors into behavior feature vectors through a recurrent neural network, wherein the behavior feature vectors include original behavior feature vectors and forwarding behavior feature vectors.
2. The method of claim 1, wherein the historical network data comprises text data, and the step of converting the historical network data into a user text feature vector comprises:
mapping the text data to a text matrix sequence based on a word vector model;
encoding the text matrix sequence into a text feature vector sequence by using a convolutional neural network;
and encoding the text feature vector sequence into a user text feature vector by using a recurrent neural network.
3. The method of claim 2, wherein the step of encoding the text matrix sequence into a text feature vector sequence using a convolutional neural network comprises:
performing convolution operation on the text matrix sequence by using the convolution neural network according to a method shown as the following formula to obtain a feature mapping matrix:
wherein S represents the text matrix sequence, FlRepresenting a filter, representing a convolution operation, m representing a width of the filter, ω representing a length of the text data, S[:,k-m+1:k]Representing an m-dimensional matrix slice, wherein k-m +1: k represents m columns of the matrix S from the k-m +1 column to the k column, and k represents an intermediate variable;
and performing pooling operation on the feature mapping matrix by using a pooling layer of the convolutional neural network to obtain a text feature vector sequence.
4. The method for social robot detection fusing user behavior and text information according to any one of claims 2-3, wherein the step of encoding the text feature vector sequence as a user text feature vector using a recurrent neural network comprises:
acquiring a user text feature vector according to a method shown in the following formula:
it＝σ(Wi[ht-1,twtt]+bi)
ft＝σ(Wf[ht-1,twtt]+bf)
qt＝tanh(Wc[ht-1,twtt]+bc)
ot＝σ(Wo[ht-1,twtt]+bo)
ct＝ft⊙ct-1+it⊙qt
ht＝ot⊙tanh(ct)
wherein itDenotes an input gate, ftIndicating a forgetting gate, qtCandidate vectors representing text data entered at time t, ctRepresenting the state of the cells of the recurrent neural network at time t, otDenotes the output gate, htIndicating an implicit state at time t, i.e. output information, Wi，Wf，Wc，Wo，bi，bf，bc，boAre learning parameters of the recurrent neural network, sigma (·) represents a sigmoid function, tanh (·) represents a hyperbolic tangent function, twttRepresenting the text feature vector.
5. The method of claim 1, wherein the step of converting the friend network data into friend network feature vectors comprises:
generating a random walk sequence corresponding to the friend network data by using a random walk algorithm;
the random walk sequence is encoded into a friend network feature vector using the Skip-Gram algorithm.
6. The method as claimed in claim 5, wherein the step of fusing the user text feature vector, the behavior feature vector and the friend network feature vector comprises:
and according to the method shown in the following formula, serially splicing the user text characteristic vector, the behavior characteristic vector and the friend network characteristic vector:
Uu＝[UCu,Pru,Rru,ctu]
wherein, UCuRepresenting said user text feature vector, PruRepresenting the feature vector of the original behavior, RruFeature vector, ct, representing forwarding behavioruRepresenting the friend network feature vector.
7. The method as claimed in claim 6, wherein the step of fusing the user text feature vector, the behavior feature vector and the friend network feature vector further comprises:
fusing the user text characteristic vector, the behavior characteristic vector and the friend network characteristic vector based on a preset weight matrix according to the method shown in the following formula:
Uu＝B+(Pru+V·Rru)+Wc·UCu+Wn·ctu
where B denotes the global bias, V denotes the weight used to balance the originating behavior vector with the forwarding behavior vector, WcWeights, W, representing the user text feature vectorsnRepresenting the buddy networkThe weight of the feature vector.
8. The method of claim 7, wherein the step of detecting the user feature vector based on a pre-constructed classification detection model comprises:
and detecting the user feature vector according to a method shown in the following formula and based on a pre-constructed classification detection model:
wherein H and H represent the weight matrix and bias of the classification detection model, UuRepresents the user feature vector, σ (-) represents a sigmoid function,
indicating the detection result.
9. A social robot detection system fusing user behavior and text information, characterized by comprising:
the acquisition module is configured to acquire historical network data and friend network data of a social media user to be detected;
a first vector conversion module configured to convert the historical network data into a user text feature vector;
a second vector conversion module configured to convert the historical network data into behavior feature vectors;
the third vector conversion module is configured to convert the friend network data into friend network feature vectors;
the fusion module is configured to fuse the user text feature vector, the behavior feature vector and the friend network feature vector to obtain a user feature vector of the to-be-detected social media user;
the classification detection module is configured to detect the user feature vector and output a detection result;
the first vector conversion module, the second vector conversion module, the third vector conversion module, the fusion module and the classification detection module are all modules which are constructed on the basis of a preset training set and by utilizing a deep neural network;
the historical network data comprises behavior data, and the second vector conversion module further comprises a behavior modeling component unit, a first coding unit, a second coding unit and a third coding unit;
the behavior modeling component unit is configured to construct an intrinsic factor behavior modeling component and an extrinsic factor behavior modeling component;
the first encoding unit is configured to encode the behavior data into an intrinsic behavior vector based on the intrinsic factor behavior modeling component;
the second encoding unit is configured to encode the intrinsic behavior vector as an extrinsic behavior vector based on the extrinsic factor behavior modeling component;
the third encoding unit is configured to encode the extrinsic behavior vectors into behavior feature vectors through a recurrent neural network, wherein the behavior feature vectors include original behavior feature vectors and forwarding behavior feature vectors.
10. The system of claim 9, wherein the historical network data comprises textual data, and the first vector transformation module further comprises a mapping unit, a convolutional neural network unit, and a recurrent neural network unit;
the mapping unit is configured to map the text data into a text matrix sequence based on a word vector model;
the convolutional neural network unit is configured to encode the text matrix sequence into a text feature vector sequence by using a convolutional neural network;
the recurrent neural network unit is configured to encode the sequence of text feature vectors into a user text feature vector using a recurrent neural network.
11. The system of claim 10, wherein the convolutional neural network unit is further configured to convolve the sequence of text matrices according to the following formula:
wherein S represents the text matrix sequence, FlRepresenting a filter, representing a convolution operation, m representing a width of the filter, ω representing a length of the text data, S[:,k-m+1:k]Represents an m-dimensional matrix slice, where k-m +1: k represents the m columns of the matrix S from the k-m +1 column to the k column, and k represents an intermediate variable.
12. The system according to any of claims 10-11, wherein the recurrent neural network element is further configured to encode the sequence of text feature vectors as user text feature vectors according to the following formula:
it＝σ(Wi[ht-1,twtt]+bi)
ft＝σ(Wf[ht-1,twtt]+bf)
qt＝tanh(Wc[ht-1,twtt]+bc)
ot＝σ(Wo[ht-1,twtt]+bo)
ct＝ft⊙ct-1+it⊙qt
ht＝ot⊙tanh(ct)
wherein itDenotes an input gate, ftIndicating a forgetting gate, qtCandidate vectors representing text data entered at time t, ctRepresenting the state of the cells of the recurrent neural network at time t, otDenotes the output gate, htIndicating an implicit state at time t, i.e. output information, Wi，Wf，Wc，Wo，bi，bf，bc，boAre learning parameters of the recurrent neural network, sigma (·) represents a sigmoid function, tanh (·) represents a hyperbolic tangent function, twttRepresenting the text feature vector.
13. The system for social robot detection with fusion of user behavior and text information according to claim 9, wherein the third vector conversion module comprises a sequence generation unit and a fourth encoding unit;
the sequence generating unit is configured to generate a random walk sequence corresponding to the friend network data by using a random walk algorithm;
the fourth encoding unit is configured to encode the random walk sequence into a friend network feature vector using a Skip-Gram algorithm.
14. The system of claim 13, wherein the fusion module further comprises a first fusion unit, and the first fusion unit is configured to concatenate the user text feature vector, the behavior feature vector, and the friend network feature vector according to the following formula:
Uu＝[UCu,Pru,Rru,ctu]
wherein, UCuRepresenting said user text feature vector, PruRepresenting the original behavior vector, RruRepresents a forwarding behavior vector, ctuRepresenting the friend network feature vector.
15. The system according to claim 14, wherein the fusion module further comprises a second fusion unit configured to fuse the user text feature vector, the behavior feature vector, and the friend network feature vector based on a preset weight matrix according to the following formula:
Uu＝B+(Pru+V·Rru)+Wc·UCu+Wn·ctu
where B denotes the global bias, V denotes the weight used to balance the originating behavior vector with the forwarding behavior vector, WcWeights, W, representing the user text feature vectorsnWeights representing the friend network feature vectors.
16. The system of claim 15, wherein the classification detection module is configured to detect the user feature vector according to the following formula:
where H and H represent the weight matrix and bias of the classification detection model, UuRepresents the user feature vector, σ (-) represents a sigmoid function,
indicating the detection result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810396753.7A CN108763319B (en) | 2018-04-28 | 2018-04-28 | Social robot detection method and system fusing user behaviors and text information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810396753.7A CN108763319B (en) | 2018-04-28 | 2018-04-28 | Social robot detection method and system fusing user behaviors and text information |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108763319A CN108763319A (en) | 2018-11-06 |
| CN108763319B true CN108763319B (en) | 2022-02-08 |
Family
ID=64012288
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810396753.7A Active CN108763319B (en) | 2018-04-28 | 2018-04-28 | Social robot detection method and system fusing user behaviors and text information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108763319B (en) |
Families Citing this family (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111401508A (en) * | 2018-12-13 | 2020-07-10 | 北京京东尚科信息技术有限公司 | Text data processing method and device and computer readable storage medium |
| CN109783614B (en) * | 2019-01-25 | 2020-11-24 | 北京信息科技大学 | Differential privacy disclosure detection method and system for to-be-published text of social network |
| CN110020431B (en) * | 2019-03-06 | 2023-07-18 | 平安科技（深圳）有限公司 | Feature extraction method and device of text information, computer equipment and storage medium |
| CN111723200A (en) * | 2019-03-20 | 2020-09-29 | 京东数字科技控股有限公司 | Method and system for determining user behavior characteristics |
| CN110070140B (en) * | 2019-04-28 | 2021-03-23 | 清华大学 | User similarity determination method and device based on multi-category information |
| CN110347797A (en) * | 2019-07-10 | 2019-10-18 | 广州市百果园信息技术有限公司 | Method for detecting, system, equipment and the storage medium of text information |
| CN110532485B (en) * | 2019-07-11 | 2022-06-03 | 中国科学院信息工程研究所 | User behavior detection method and device based on multi-source data fusion |
| CN110457404B (en) * | 2019-08-19 | 2021-11-02 | 电子科技大学 | Social media account classification method based on complex heterogeneous network |
| CN110689164B (en) * | 2019-08-26 | 2023-04-28 | 创新先进技术有限公司 | Prediction method and system for user offer behavior |
| CN110633423B (en) * | 2019-09-20 | 2022-09-09 | 北京字节跳动网络技术有限公司 | Target account identification method, device, equipment and storage medium |
| CN110781407A (en) * | 2019-10-21 | 2020-02-11 | 腾讯科技（深圳）有限公司 | User label generation method and device and computer readable storage medium |
| CN110837602B (en) * | 2019-11-05 | 2022-10-04 | 重庆邮电大学 | User recommendation method based on representation learning and multi-mode convolutional neural network |
| CN111340112B (en) * | 2020-02-26 | 2023-09-26 | 腾讯科技（深圳）有限公司 | Classification method, classification device and classification server |
| CN111428116B (en) * | 2020-06-08 | 2021-01-12 | 四川大学 | Microblog social robot detection method based on deep neural network |
| CN112395511A (en) * | 2020-11-04 | 2021-02-23 | 北京大学 | Rumor detection and evidence preservation system based on message propagation path in mobile application |
| CN112700342B (en) * | 2020-12-29 | 2022-09-13 | 四川大学 | Personalized grading prediction model and prediction method for social media information authenticity |
| CN112685614B (en) * | 2021-03-17 | 2021-06-18 | 中国电子科技集团公司第三十研究所 | Social media robot group rapid detection method |
| CN113705099B (en) * | 2021-05-09 | 2023-06-13 | 电子科技大学 | Social platform rumor detection model construction method and detection method based on contrast learning |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102571484A (en) * | 2011-12-14 | 2012-07-11 | 上海交通大学 | Method for detecting and finding online water army |
| CN103077240A (en) * | 2013-01-10 | 2013-05-01 | 北京工商大学 | Microblog water army identifying method based on probabilistic graphical model |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103955714A (en) * | 2014-04-09 | 2014-07-30 | 中国科学院信息工程研究所 | Navy detection model construction method and system and navy detection method |
| US20160350675A1 (en) * | 2015-06-01 | 2016-12-01 | Facebook, Inc. | Systems and methods to identify objectionable content |
-
2018
- 2018-04-28 CN CN201810396753.7A patent/CN108763319B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102571484A (en) * | 2011-12-14 | 2012-07-11 | 上海交通大学 | Method for detecting and finding online water army |
| CN103077240A (en) * | 2013-01-10 | 2013-05-01 | 北京工商大学 | Microblog water army identifying method based on probabilistic graphical model |
Non-Patent Citations (1)
| Title |
|---|
| 恶意社交机器人检测技术研究;刘蓉等;《通信学报》;20171130;第38卷(第Z2期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108763319A (en) | 2018-11-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108763319B (en) | Social robot detection method and system fusing user behaviors and text information | |
| CN112199608B (en) | Social media rumor detection method based on network information propagation graph modeling | |
| CN113239186B (en) | Graph convolution network relation extraction method based on multi-dependency relation representation mechanism | |
| Remesan et al. | Hydrological data driven modelling | |
| CN111008337B (en) | Deep attention rumor identification method and device based on ternary characteristics | |
| Ray et al. | A backcast land use change model to generate past land use maps: application and validation at the Muskegon River watershed of Michigan, USA | |
| CN110147911B (en) | Social influence prediction model and prediction method based on content perception | |
| CN110336838B (en) | Account abnormity detection method, device, terminal and storage medium | |
| CN110378124A (en) | A kind of network security threats analysis method and system based on LDA machine learning | |
| CN111382190A (en) | Object recommendation method and device based on intelligence and storage medium | |
| CN112561031A (en) | Model searching method and device based on artificial intelligence and electronic equipment | |
| CN111428151B (en) | False message identification method and device based on network acceleration | |
| CN117061322A (en) | Internet of things flow pool management method and system | |
| CN112529638A (en) | Service demand dynamic prediction method and system based on user classification and deep learning | |
| CN114329455B (en) | User abnormal behavior detection method and device based on heterogeneous graph embedding | |
| CN111198897A (en) | Scientific research hotspot topic analysis method and device and electronic equipment | |
| Casleton et al. | A local structure model for network analysis | |
| Dutta et al. | Big data architecture for environmental analytics | |
| CN112712855A (en) | Joint training-based clustering method for gene microarray containing deletion value | |
| CN114265954B (en) | Graph representation learning method based on position and structure information | |
| CN115099344A (en) | Model training method and device, user portrait generation method and device, and equipment | |
| CN115841334A (en) | Abnormal account identification method and device, electronic equipment and storage medium | |
| CN116091133A (en) | Target object attribute identification method, device and storage medium | |
| CN107767278B (en) | Method and device for constructing community hierarchy | |
| CN114943016A (en) | Cross-granularity joint training-based graph comparison representation learning method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |